purplecat: Hand Drawn picture of a Toy Cat (aisb)
Prompted because I was just sent a link to his obituary to put on the AISB website.
purplecat: Hand Drawn picture of a Toy Cat (aisb)
Luciano Floridi gave two invited talks at the AISB convention. The first was a two-handed public lecture with Aaron Sloman. Aaron's talk was broadly similar to his recent Thinking about Mathematics and Science lecture at Liverpool. The second was an invited talk for the academics at the conference but Floridi treated them as two halves of the same thing.

Floridi is a primarily a philosopher. His interest, as I understood it, is in understanding philosophically what is happening at the moment in the interaction between humans and computational systems, in particular with a hope that this will allow us to avoid pitfalls down the road. He made a number of interesting points which I'm going to cover in no particular order:


  • We are on the edge of a shift in how we view ourselves; "The fourth revoluation". Once we thought we were the centre of the universe but then we had to change that self perception (The Copernican Revolution). Then we thought we were uniquely created and had to change that (The Darwinian Revolution). Then we viewed ourselves as entirely rational and explicable organisms (Freud put a stop to that one). I wasn't entirely clear exactly what change in self-perception the fourth revolution was but I think it involved challenging our perception of ourselves as discrete physical objects in favour of one that viewed ourselves as interconnected informational objects. There was a surprisingly vehement negative response to this idea in much of the room (though that response was linked to my next point) which suggested that, at the very least, the concept does challenge people's perception of self in some way.

  • We are a long way from producing intelligent programs but we already have a lot of dumb but smart systems. For instance Neopets are very basic but nevertheless clearly fill an emotional need for a lot of people. Floridi posited an upsurge of dumb programs designed to mimic human companionship in very specific ways - some of these would be for entertainment only (like Neopets) but some would have more specific assistive functions (e.g., monitoring of the elderly). None would be anything like intelligent. At lot of discussion followed on whether people would be "fooled" by this. Further discussion followed that people wouldn't be "fooled" - they'd be quite aware of the limitations of such companions - but they would use them and become attached to them anyway just as they do to pets or, perhaps more relevantly, sentimental objects.

  • The Ancient Greeks had an animist view of the world in which all objects had, to some extent, a personality. With the advent of pervasive systems and RFID tags making it practical to embed limited interactivity into everyday objects we might well be cyclicly entering a view of the world in which objects once more have personality (or at least a form of interactivity). Right now its only cars that talk back to us (and only if we have GPS installed).

  • At the moment most of us view the online/informational world as, in some sense, separate from the real or physical world. As pervasive systems become more widespread this concept of separation will fade and we will less and less compartmentalise what we are doing as either an informational act (working at a computer) or a physical act (not working at a computer).

purplecat: Hand Drawn picture of a Toy Cat (aisb)
You all thought I'd finished didn't you! I was just pausing. The last slew of talk posts were all for Thursday. This is a Friday talk but I actually have less to blog about on Friday. I heard fewer interesting talks, and several of those I did hear, I had heard before. Ethan Kennerly of Fine Game Design was invited to talk about, you know, actual games rather than game theory ones. He was particularly interested in the connection between the game mechanics (or simulation as he called it) and the game story. So, for instance, in chess the simulation (rules) is embedded loosely in a battle/combat story. Obviously in some games the story aspect is more important than in others. Kennerly attempted to highlight some practical problems in game design, particularly those related to making the simulation and story match up in an aesthetically pleasing fashion which he felt might benefit from some sort of theoretical framework which would help game designers understand what they were doing. The areas he highlighted were:


  • Correlating the simulation dynamics of risk and return to the aesthetic experience of play - so when a player does something risky in game mechanical terms they should understand it as risky on the story level.

  • Developing a theoretical relationship between challenge construction, skill acquisition and the aesthetics of drama

  • How does the combinatorial game theoretic heat of a simulation state correlate to the aesthetic experience of its users?

  • How do knowledge games (e.g., ones of bluff and information acquisition) extend to model correlations of a user's aesthetic experience?

  • How to make the dynamics of the simulation and story channel in a dramatic game reveal character and advance conflict toward a conclusion



The presentation was a lot of fun as well as being interesting. Kennerly kept getting us to suggest either story choices that might match a particular simulation or a simulation that might match a particular story.

He also brought with him a card game he'd devised for his son to help him learn mental arithmetic. It was a surprisingly complex to play, especially in a version where you played as partners, and had us all sitting around trying to do subtraction and multiplication in our heads.

On Friday evening four of us, including Kennerly, went out for dinner during which it became clear that three of us were roleplayers. We then went on to have one of those conversations which is very boring if you happen to be the fourth person at the table - though he asked for it somewhat by asking us to explain what roleplaying actually was, which devolved into an argument about whether one-off freeform political type games counted as roleplaying; how character and mechanics should be balanced; and the extent to which the player's skills determined the character's skills. Not to mention the aesthetics of rolling lots of dice at once in order to simulate a really big fireball.
purplecat: Hand Drawn picture of a Toy Cat (aisb)
All I'm going to note about this talk is that I find the iCat more disturbing than cute.

I sat next to the speaker, Frank Dignum, at dinner that evening and, as well as being very nice, he said some very perspicacious things about organisations of agents which I'm going to have to think about.
purplecat: Hand Drawn picture of a Toy Cat (aisb)
Maarten Schadd (with co-authors Mark Winands, Jaap van den Herik and Huib Aldewereld) gave a talk whose primary interest, from my POV, was that the bricks breaking game on Facebook is NP-Complete.

I'm going to have to explain that aren't I.

A P-time puzzle is one which, to all intents and purposes, can be solved quickly (according to a technical definition of quick). An NP-time puzzle is one in which, if you have the right answer, you can check it is right quickly but you can't necessarily find the right answer quickly. No one knows if P=NP though most people suspect not. Field medals will be won and a lot of research will get torn up if it turns out that P does equal NP.

I rather like bricks breaking.
purplecat: Hand Drawn picture of a Toy Cat (aisb)
This is the last experimental game theory talk I'm going to precis - but it was a fascinating session and an area I'd not come across before.

This talk was by local man, Juergen Bracht. The "trust game" is one which involves an investor and an allocator. The investor starts with 2 points and can keep them or invest them. If (s)he invests them they automatically grow to 8 pts which the allocator can then either keep entirely for themselves or split between the investor and the allocator. Juegen was interested in the effects of two processes on the trust game. The first, "cheap talk", was where the allocator was allowed to tell the investor what they intended to do with the points in advance, but was not held to that utterance. The second, "observation", was where the investor had access to the allocator's previous actions.

Unsurprisingly, especially since the interaction was computer mediated, not face-to-face, cheap talk had little effect on investor or allocator actions. Less surprisingly observation did have an effect (except in the last round - where both investor and allocator knew it was the last round). The reason I say "less surprisingly" is became in pure game theoretic terms the reasoning goes: in the last round the allocator should keep all the points since no one will ever use the outcome of the round in an observation, therefore the investor should not invest their points. This being the case the second-to-last round is the last one where anyone will invest so the allocator should keep all the points since no one will ever use the outcome of the round in an observation, therefore the investor should not invest their points. This being the case... and so on so no one ever invests anything... Clearly classical game theory needs some rethinking if we expect it to realistically model human actions.
purplecat: Hand Drawn picture of a Toy Cat (aisb)
Sobei H. Oda (together with co-authors, Gen Masumoto and Hiroyasu Yoneda) had been simulating the (I think) Future's market. In particular he had been testing the hypothesis that you make more money if you have better information about the future price of a commodity. He had a graph that showed the people with no information doing a little better than those with just a little information. The people with lots of information still did best. He had some maths to explain this, which it must be said I didn't follow, but I thought it was an interesting effect.
purplecat: Hand Drawn picture of a Toy Cat (aisb)
Vincent Wiegel presented joint work with Jan van den Berg investigating a criticism of a philosophical standpoint called act utilitarianism (contrasted to rule utilitarianism).

Roughly speaking, an act utilitarian evaluates each action, as they occur, in order to decide the utility of acting while a rule utilitarian acts according to a general rule about the utility. The thought experiment used to debunk act utilitarianism was that of an election. In a population of 100 an act utilitarian only votes if they are the 51st person to vote for their preferred candidate in all other situations they gain more utility by going and doing something else more interesting. Wiegel and van den Berg simulated this situation computationally. Obviously first they had some issues about why an act utilitarian might conclude they get utility only by being the 51st person to vote and of course, how they might determine that they have the deciding vote. Interestingly, when they varied their assumptions a bit so that act utilitarians only voted if they had reason to believe they were in the range of the 46th - 56th voter (or similar) - i.e., that their vote was likely to decisive then they did very well frequently getting the outcome they wished in an election while getting to do other more interesting things when the outcome was essentially a foregone conclusion anyway.
purplecat: Hand Drawn picture of a Toy Cat (aisb)
I've just got back from the AISB Convention in Aberdeen. I'm just about up-to-date on reading email and LJ but a long way behind in terms of responding. There are also lots of things I want to blog about including several really interesting talks. I was going to do one big "interesting talks" post but that was intimidating so I thought I'd do lots of short ones.

The first is Justine Cassell's talk on Virtual Peers for Studying and Scaffolding Real Social Interaction. Justine had studied interaction, specifically storytelling, in children and created a "Virtual Peer" which would then interact with a child. This interaction was via a doll's house (half real half virtual). Certain parts of the Doll's House were covered and a child would put a doll behind one of these covers to let the virtual child use it (apparently only AI researchers try to break this mechanism and have one doll in two places). The working of the Virtual Peer involves Justine sitting in a cupboard with a control panel directing its actions and responses. One of the really interesting bits was when they extended this work to autistic children and compared their interactions with the Virtual Peer to their interactions with a real child. The autistic child responded more often and participated more in the joint story-telling when interacting with the Virtual Peer than with a real child - they have some hypotheses why this might be (some of which are testable and some of which aren't - or at least they can't think how yet). They are extending the work to allow children to construct and operate the control panel themselves and are hoping this will enable autistic children to "experiment" with social interaction.
purplecat: Hand Drawn picture of a Toy Cat (aisb)
Title: With powerful languages and good programming

Contents: more valuable, is that Lisp programs are expressed as Lisp data

This got sent from a yahoo address to a now defunct vicechair email address for the AISB (which somehow got filtered to me rather than being bounced) and Lisp is the first AI language. There was nothing else discernable in the email in terms of attachments, links or offers to procure a russian wife.... on the other hand its not as if it exactly makes much sense as a communication.

Since it seems to refer to Lisp's ability to recursively process its own code perhaps its some secret terrorist plot being transmitted by email: Recursion. If you suspect it, report it.

Profile

purplecat: Hand Drawn picture of a Toy Cat (Default)
purplecat

October 2017

S M T W T F S
123456 7
8910111213 14
15161718192021
22232425262728
293031    

Syndicate

RSS Atom

Tags

Style Credit

Expand Cut Tags

No cut tags